All work

Design studio · 2026

Pre-sales Agent

A pre-sales agent in Notion for a design studio. Its estimates start from the studio's templates for six project types, and a project manager approves each step.

Role
Agent design, prompt, checklists, brief template, client questionnaire
Period
2026
Status
Designed and tested on five sample inquiries.
Stack
  • Notion custom AI agent
  • Gmail and Google Drive connectors
  • Notion knowledge base
On this page
  1. 01Why
  2. 02How it fits together
  3. 03The estimate
  4. 04Making it checkable
  5. 05Testing
Pre-sales agentNovotný & Partners · new website
  1. 1Analysis
  2. 2Scope
  3. 3Brief
  4. 4Estimate
Clickable demo on a sample inquiry. It runs without a model, and step one follows a test run of the agent.

Why

Past projects showed where the estimates missed. Project management was underestimated, mobile UI took far longer than planned, and feedback rounds had no limit. A generic chatbot doesn’t know that, or what the studio needs from a client before work starts.

How it fits together

Scope waits for the client’s answers to step one. I also put the main lessons into the prompt, so they apply on every run, not only when the agent looks them up.

The estimate

The estimate is a draft for the project manager, not a quote. I wrote the rules it follows.

  • It starts from the studio’s template for the closest of six project types. The agent changes the counts to fit the approved scope and says why for each changed row.
  • Project management is a formula, a base plus a percentage, so it can’t be squeezed.
  • A range, not one number. The upper end covers mobile UI, where past projects ran over most.
  • If the price is too high, it cuts scope, never project management or the two feedback rounds.

Making it checkable

  • Scoped access. The agent reads only this inquiry’s emails and folder, and the knowledge base.
  • A source on every point, down to the checklist item.
  • Gaps stay visible. Missing inputs become questions for the client, not assumptions.
  • Confidence from fixed rules: which of budget, deadline, audience and scope are known.

Testing

I ran it on five made-up inquiries with problems planted on purpose, like a website brief that hides an app with a client login. Step one of the demo comes from one of those runs.